How to Conduct a Systematic Review: A Complete Step-by-Step Guide

What Is a Systematic Review?

A systematic review is a rigorous, reproducible method for identifying, evaluating, and synthesizing all relevant studies on a specific research question. It follows a pre-defined protocol to minimize bias and represents the highest level of evidence in evidence-based medicine.

Unlike narrative reviews, systematic reviews emphasize transparency, reproducibility, and comprehensiveness, providing reliable evidence for clinical decision-making and guideline development.

Step 1: Define Your Research Question (PICO Framework)

A well-defined question is the foundation of any systematic review. Use the PICO framework:

P (Population): Who are the participants? e.g., adults with type 2 diabetes
I (Intervention): What is the intervention? e.g., GLP-1 receptor agonists
C (Comparison): What is the comparator? e.g., placebo or conventional therapy
O (Outcome): What outcomes matter? e.g., HbA1c change, weight change, adverse events

Step 2: Develop a Search Strategy

The search strategy directly determines the comprehensiveness of your review:

Multiple databases: Search at least PubMed, Embase, and Cochrane Library
Build search queries: Combine MeSH terms and free-text keywords using Boolean operators (AND/OR)
No language restrictions: Avoid language filters to reduce publication bias
Supplementary searches: Check reference lists, grey literature, and trial registries (ClinicalTrials.gov)

Have an information specialist review your strategy. ResearchPilot can help you quickly generate and refine search queries with AI.

Step 3: Study Selection (PRISMA Flow)

Follow the PRISMA 2020 guidelines for a two-stage screening process:

Title/Abstract screening: Two independent reviewers screen records against inclusion/exclusion criteria
Full-text screening: Retrieve and assess full texts of potentially eligible studies

Document exclusion reasons at each stage and create a PRISMA flow diagram. Assess inter-rater agreement using Cohen's kappa and resolve disagreements through discussion or a third reviewer.

Step 4: Data Extraction

Design a standardized extraction form covering:

Study characteristics: Authors, year, country, study design
Participant details: Sample size, age, sex, baseline characteristics
Intervention and comparator: Regimen, dosage, duration
Outcome data: Means, standard deviations, event counts, totals
Risk of bias items: Randomization method, blinding, attrition rate

Two reviewers should extract data independently and cross-check. Contact original authors for missing data when possible.

Step 5: Risk of Bias Assessment

Choose the appropriate tool based on study design:

RCTs: Cochrane RoB 2 tool (5 domains: randomization, deviations, missing data, measurement, selective reporting)
Non-randomized studies: ROBINS-I tool
Observational studies: Newcastle-Ottawa Scale (NOS)

Present results using traffic light plots and summary plots. Consider excluding high-risk studies in sensitivity analyses to test robustness.

Step 6: Data Synthesis and Meta-Analysis

If clinical and methodological heterogeneity is acceptable, perform quantitative synthesis:

Effect measures: Mean difference (MD) or standardized mean difference (SMD) for continuous outcomes; risk ratio (RR) or odds ratio (OR) for binary outcomes
Model selection: Fixed-effect (low heterogeneity) or random-effects model (more common)
Heterogeneity assessment: I-squared statistic (above 50% suggests substantial heterogeneity), Cochran's Q test
Forest plot: Visualize individual and pooled effect estimates

Also conduct subgroup analyses (explore heterogeneity sources), sensitivity analyses (test robustness), and publication bias assessment (funnel plot + Egger's test).

Step 7: Evidence Quality (GRADE)

The GRADE framework evaluates certainty of evidence across five domains:

Risk of bias
Inconsistency (heterogeneity)
Indirectness
Imprecision
Publication bias

Evidence is rated as high, moderate, low, or very low. Present GRADE results in a Summary of Findings (SoF) table, one of the most important outputs of a systematic review.

Step 8: Writing and Publication

Follow the PRISMA 2020 statement (27 checklist items) when writing. Pre-submission checklist:

PROSPERO registration number included
PRISMA flow diagram complete
Search strategy reproducible (full queries provided)
Risk of bias assessment results complete
Forest plots, funnel plots, and key figures included
GRADE Summary of Findings table prepared
Limitations thoroughly discussed

Target journals: Cochrane Database of Systematic Reviews, BMJ, JAMA, or specialty journals that welcome systematic reviews.

How AI Is Accelerating Systematic Reviews

Traditional systematic reviews take 6-18 months. AI tools are changing this:

Smart search: AI-assisted search strategy generation and optimization
Automated screening: AI pre-screening based on inclusion/exclusion criteria
Data extraction: Automatic identification and extraction of key data from full texts
Bias assessment: AI-assisted RoB 2 evaluation with preliminary judgments

ResearchPilot's MetaScholar feature aims to compress the entire systematic review workflow from months to days while maintaining methodological rigor. From literature search and PRISMA screening to data extraction and statistical analysis, all in one place.

Ready to experience AI-powered systematic reviews? Try ResearchPilot today and make your research more efficient.